KT and Samsung cut KTX communication dropouts by 29% using AI

KT and Samsung Electronics have cut communication dropouts on Korean high-speed trains by roughly 29%, using artificial intelligence to change how and when a moving train hands its connection from one base station to the next. KT announced the completed trial on November 5.

The result comes from measured data, not projection. Comparing before-and-after figures on the test section, the two companies found that 100 dropouts previously occurring would fall to about 71 with the technology applied. Passengers on KTX and SRT services have long faced dropped calls and interrupted data sessions in high-speed running, and the trial addresses the mechanism behind those failures directly.

Where and how the trial ran

The trial took place on a stretch of the Gyeongbu Line high-speed railway between Godeok-myeon in Pyeongtaek City and Maesong-myeon in Hwaseong City. The corridor is part of Korea's principal north-south high-speed axis, where trains run at 250 km/h or faster. At those speeds a train's serving base station changes frequently, and any delay in the handover — the transfer of an active connection from one station to the next — produces temporary disruption to data transmission or phone calls.

KT identified handover delay as the dominant cause of dropouts: trains moved faster than the network could switch them to the next station. The fix was to bring forward the timing of handovers at sections where failures recurred most often, so the connection transfers before it would otherwise be lost.

What changed in the engineering method

The method itself shifted as much as the outcome. Previously, engineers inspected individual base stations showing degraded service quality through on-site measurements — a station-by-station approach. In this trial, the two companies instead analyzed large-scale communications data: signal strength records and connection-failure logs collected along high-speed rail sections. The AI processed that data to identify where failures recur and what patterns govern how connections are made.

That analytical shift matters for cost. Finding problem sections in bulk data, rather than dispatching crews to measure each suspect station, targets engineering effort at the locations where it pays. It also produced the specific intervention — earlier handovers — that the trial then validated in the field.

What does this change for passengers and the network?

For passengers, the measured 29% reduction means fewer dropped calls and fewer interruptions to data sessions on the test section. For the network, it demonstrates that radio-layer optimization can be tuned to the dynamics of 250 km/h operation, where conventional per-station maintenance struggles to keep pace. The gain is currently confined to the trial corridor, and KT has committed only to consider implementing the technology on high-speed rail sections based on the trial results — no rollout date, no coverage scope, and no further performance targets have been announced.

The wider program continues regardless. KT and Samsung Electronics plan to keep developing and validating AI-based radio access network technology, known as AI-RAN (the application of AI to the part of the mobile network that connects devices by radio), as a joint effort.

Executive commitments

Jin-guk Jeong, vice president and head of the Next Generation Communications Research Center at Samsung Research, Samsung Electronics, said: "We will continue to take the lead together with KT in core AI-RAN technologies for user-centered experience innovation."

Jongsik Lee, head of the Future Network Lab in KT's Network Division, said: "We will continue developing AI-RAN technology to bring about tangible improvements in the quality customers experience."

Both statements are forward-looking claims without attached deliverables or dates; neither company has published a schedule for extending the handover-timing technique beyond the Gyeongbu Line test section. What exists today is a validated trial result on one corridor, a method that scales through data rather than site visits, and a stated intention to apply it across Korea's high-speed network.